(ECNS) -- Chinese artificial intelligence startup Zhipu AI on Monday unveiled a free AI agent, which can not only perform deep research but also execute tasks such as web searches, travel planning, and research report writing.
The AI agent, named AutoGLM Rumination, is powered by Zhipu's proprietary models, including its reasoning model GLM-Z1-Air and foundation model GLM-4-Air-0414. The core chain of models and technologies is scheduled for full open-source release on April 14, the company said.
AutoGLM Rumination relies on three key features: deep thinking, perception of the world, and use of tools.
In contrast to some of the leading international technologies, AutoGLM Rumination is not only capable of conducting in-depth research, but also of actually executing tasks, thus propelling the AI Agent from a mere thinker to an intelligent performer capable of delivering results, according to the company.
Through reinforcement learning, the model powered AI agent is capable of self-criticism, reflection, and even deep thinking, thus developing the ability for long-range reasoning and task execution.
Currently, the AI agent is available on the PC edition of ChatGLM where users can experience its research and operational capabilities for free. This release is a preview version, which mainly supports research scenarios.
The AutoGLM Rumination research team will further expand the execution ability of the intelligent body, and launch a "virtual machine" version for enhanced scenarios, the company said.
Zhipu stated that this marks the first publicly available Deep Research functionality in China's domestic AI market.
The company, founded in 2019 as a spinoff from a Tsinghua University laboratory, has emerged as one of China's leading AI startups. Having developed the GLM series of models, Zhipu claims its latest large language model GLM4 outperforms OpenAI's GPT-4 on several benchmarks.
(By Zhang Dongfang)
京公网安备 11010202009201号